Everybody's System, Nobody's Risk

When an AI-supported decision goes wrong, the supplier points to its terms, the application builder to an unannounced model update, the professional to the tool, and each regulator to its own remit. Everyone participated and nobody owned the outcome. This paper argues that the fragmentation requires no one to behave badly, being the predictable result of distributing responsibility without anyone holding the whole picture, and that learning systems, shared foundation models, informal staff use and deliberately distributed regulatory oversight all deepen it. It maps four links in a generative AI supply chain against what each controls, what each can see and what each carries, and identifies the mismatch between the last two columns as the accountability gap in commercial form. A fifth link, staff using tools nobody approved, sits outside the chain entirely. The remedy proposed is available without waiting for regulators: one named owner for every AI-supported decision, a responsibility map redrawn whenever a supplier changes its model or terms, and internal audit that tests the seams between parties. The owner model is the author's own and has not been tested at scale.